Fury X 3DMark 13 Mega bench (FS / FSE / FSU over 250+ runs) by gupsterg - Overclock.net - An Overclocking Community

Forum Jump: 

Fury X 3DMark 13 Mega bench (FS / FSE / FSU over 250+ runs) by gupsterg

Reply
 
Thread Tools
post #1 of 19 (permalink) Old 01-24-2017, 01:59 PM - Thread Starter
Meddling user
 
gupsterg's Avatar
 
Join Date: Jan 2015
Location: Lurking over a keyboard
Posts: 6,931
Rep: 739 (Unique: 344)
Under construction, check back soon, last update 26/01/17 smile.gif .

This thread is based on data I collected for a Sapphire R9 Fury X. This card has ASIC quality of 70.4%, this is LeakageID of GPU, the OP of Hawaii bios mod has heading What is "ASIC Quality"? with information collated from The Stilt's posts. I have also been comparing ASIC quality/VID per DPM data for the cards I have owned, some of this information is in this linked post.

Fury X no 3 has ASIC Quality 64.4% and stock DPM 7 VID of 1212mV, Fury X no 8 ASIC quality of 70.4% and stock DPM 7 VID of 1200mV. Fury X no 3 has for months been used at GPU: [email protected] (ie +68mV over stock) HBM: 545MHz 1325mV (ie +25mV over stock). This usage has included synthetic bench runs & lengthy loops/gaming/[email protected]

Fury X no 8 only achieved [email protected] fully stable IMO (ie ~3hrs 3DM FS/Valley/Heaven loops, 24hrs+ continuous [email protected]). Although was very capable on just bench runs for higher clocks, which other cards have not been. All the bench runs were done back to back in stages, IIRC over ~3 days I managed the runs. The HBM seemed comfortable at 545MHz even with stock MVDDC, due to time limitations I did not give it the same gruelling testing as Fury X no 3 for HBM clock stability. HBM was pretty decent on Fury X no 8, as it's the first card out of 11 Fiji cards I've owned which allowed 600MHz HBM clock for bench testing (+100mV over stock). For bench tests I used +56mV over stock (1356mV) for HBM clock 545MHz and +100mV over stock (1400mV) for HBM clock 600MHz.

Rig:- Post screenie of CPU-Z / GPU-Z , etc

Note: Ignore driver/bios info in GPU-Z screenie below, these are historic screenies when meddling with cards, the purpose of screenie is to show ASIC quality and VID per DPM.





OS setup:- Win 10 Pro x64 (Power Profile Performance so CPU stick at max clocks), Crimson Relive v16.12.2 WHQL (Driver is default settings but FreeSync Off, Power Efficiency Off).

Below is attached a HML file from MSI AB, this has monitoring data from 3DM combined test loop for ~5mins, room ambient temp ~24°C.


3DM_HML.zip 11k .zip file

Below is HWiNFO screenie carried out whilst above run was done.





Info from HWiNFO in regard to GPU A/W is wrong, for example GPU current max is shown as 278A, as OCP in IR3567B is 240A GPU can not use that. This is not an issue with HWiNFO but how the driver is seeing data which it displays.

IR3567B measures the power / current flowing over the inductors (DCR), this is highlighted in IR3565B datasheet, so I regard occurrence of OCP violation on Fury X no 8 as accurate measure compared with what driver is passing to HWiNFO.

Notes:

(i) Total system draw was measured from wall socket. System idle wattage was MAX 95W. MAX figure given in benchs below is when 3DM combined test was occurring (ie CPU/GPU loaded, this was highest figure out of 3DM full run).

(ii) Fury X no 8 when set to 1050MHz 1262mV tripped OCP of 216A on IR3567B, had to make stock 240A. Later testing showed 228A was ample. In comparison, Fury X no 3 can use 216A OCP for 1145MHz @ 1268mV, at 1175MHz @ 1268mV it needs higher OCP, no clocks between 1145MHz and 1175MHz tested yet for this aspect.

(iii) PowerLimit was expanded in ROM, stock is TDP: 270W TDC: 300A MPDL: 300W, expanded values TDP: 350W TDC: 325A MPDL: 350W

Guide to customising runs links (Click to show)
My 3DM FS / FSE / FSU runs are hidden on FM DB, so the only way they can be viewed is as compares. These compares can be customised down to 1 run. The format of the link must be:-
Code:
Link format example 1x result: http://www.3dmark.com/compare/fs/<insert result ID>

Link example 1x result: http://www.3dmark.com/compare/fs/11453408

Link format example 2x result: http://www.3dmark.com/compare/fs/<insert result ID>/fs/<insert result ID>

Link example 2x result: http://www.3dmark.com/compare/fs/11453408/fs/11453439

Link format example 3x result: http://www.3dmark.com/compare/fs/<insert result ID>/fs/<insert result ID>/fs/<insert result ID>

Link example 3x result: http://www.3dmark.com/compare/fs/11453408/fs/11453439/fs/11453473

So let's say you wish to compare one of your own results against a posted link, you can find your result ID on your run in details section. Then just either replace a result ID of mine in my posted links with your result ID or add it to my link as shown above in example.



3DM FS runs links (Click to show)
Fury X no 8 3DM FS runs using HBM 500MHz with various GPU clocks.

Warning: Spoiler! (Click to show)

Fury X no 8 3DM FS runs using HBM 545MHz with various GPU clocks.

Warning: Spoiler! (Click to show)

Fury X no 8 3DM FS runs using HBM 600MHz with various GPU clocks.

Warning: Spoiler! (Click to show)

3DM FSE runs links (Click to show)
Fury X no 8 3DM FSE runs using HBM 500MHz with various GPU clocks.

Warning: Spoiler! (Click to show)

Fury X no 8 3DM FSE runs using HBM 545MHz with various GPU clocks.

Warning: Spoiler! (Click to show)

Fury X no 8 3DM FSE runs using HBM 600MHz with various GPU clocks.

Warning: Spoiler! (Click to show)

3DM FSU runs links (Click to show)
Fury X no 8 3DM FSU runs using HBM 500MHz with various GPU clocks.

Warning: Spoiler! (Click to show)

Fury X no 8 3DM FSU runs using HBM 545MHz with various GPU clocks.

Warning: Spoiler! (Click to show)

Fury X no 8 3DM FSU runs using HBM 600MHz with various GPU clocks.

Warning: Spoiler! (Click to show)

3DM TS to be posted soon, so check back smile.gif . Soon there will be sections comparing negative performance scaling on Fiji with voltage increase, etc, etc wink.gif .
gupsterg is online now  
Sponsored Links
Advertisement
 
post #2 of 19 (permalink) Old 01-24-2017, 02:09 PM
New to Overclock.net
 
ducegt's Avatar
 
Join Date: Jul 2015
Posts: 750
Rep: 36 (Unique: 29)
Madness! Very interesting to see such variation knowing it's the actual cards and not differences that come from various methodologies and systems that vary when multiple people make comparisms.

ducegt is offline  
post #3 of 19 (permalink) Old 01-24-2017, 05:41 PM - Thread Starter
Meddling user
 
gupsterg's Avatar
 
Join Date: Jan 2015
Location: Lurking over a keyboard
Posts: 6,931
Rep: 739 (Unique: 344)
Fiji is quite different AFAIK in how it behaves with voltage increase.

What happens with Fiji is extra voltage will lead to performance drop. I now believe it is not happening at x point but all the time. View these 3 runs each of 1137mV (undervolt) vs 1200mV (stock VID) vs 1262mV (overvolt) , consistently undervolt is best for performance disregarding power draw.

This effect occurs the same regardless of ASIC quality.

Soon there will be benchs of Fury X no 3 to compare with no 8 smile.gif .
gupsterg is online now  
Sponsored Links
Advertisement
 
post #4 of 19 (permalink) Old 01-25-2017, 05:26 AM
New to Overclock.net
 
LionS7's Avatar
 
Join Date: Dec 2012
Location: Bulgaria
Posts: 129
Rep: 1 (Unique: 1)
Those test are no good. Need to test Battlefront / Battlefield 1 for around 2h, no vsync, 100% gpu load.. Timespy stress test is good for HBM only. The difference from Unigine/3Dmark and Frostbite 3 is - 1.23V to 1.26V+ for me.
LionS7 is offline  
post #5 of 19 (permalink) Old 01-25-2017, 11:10 AM - Thread Starter
Meddling user
 
gupsterg's Avatar
 
Join Date: Jan 2015
Location: Lurking over a keyboard
Posts: 6,931
Rep: 739 (Unique: 344)
Quote:
Originally Posted by LionS7 View Post

Those test are no good. Need to test Battlefront / Battlefield 1 for around 2h, no vsync, 100% gpu load..

I don't have BF1, I do have SWBF wink.gif , but I'm not gonna spend 2hrs for each test scenario! perhaps you can wink.gif .

There are 9x GPU clocks/voltages I used, then 3x HBM clocks = 27 benches, at 2hrs each I'd be dedicating 54hrs as per your testing.

Why are these tests no good? please explain smile.gifheadscratch.gifthinking.gif.

For me I see nothing wrong in the testing smile.gif . Anyone can do a 3DM run quickly and easily and compare, data can be shared easily for comparing using the FM DB link.

I have been able to see that I experience drop on scaling at every stage of voltage increase. I was not too surprised that undervolting consistently yielded better results than same clocks at stock/increased voltage. I've also seen higher ASIC quality didn't yield better scaling in a way for performance, it also has same negative scaling with voltage increase, it also needed higher OCP vs lower ASIC quality GPU.
Quote:
Originally Posted by LionS7 View Post

Timespy stress test is good for HBM only.

TS is not just good for HBM wink.gif , it is also affected by GPU clock (and CPU) wink.gif , there will be some TS benchs added soon smile.gif . Some will be comparing Fury X no 3 vs no 8 wink.gif , then there is something I mentioned in Fury X owners club regarding what I saw when doing TS runs which will be shown here also for further discussion perhaps? wink.gif .
Quote:
Originally Posted by LionS7 View Post

The difference from Unigine/3Dmark and Frostbite 3 is - 1.23V to 1.26V+ for me.

And this has what to do with my testing? smile.gifheadscratch.gifthinking.gif .

I am not stating for every situation power draw will be same wink.gif .

Nor am I stating VDDC will be the same for each situation of GPU loading (ie 3DM/Valley/Heaven,etc) wink.gif .

You may have noted I stated VID (ie what the GPU request/set to) wink.gif . So the variability of LLC will be the same as each test is same GPU load smile.gif , yes the phenomenon of LLC will differ when comparing Fury X no 3 and no 8 as LeakageID differs smile.gif , but that will differ when others compare their GPU results with anothers wink.gif .

I think I will add a section in OP regarding why VID is stated and not just I used +56mV and so on wink.gif .
gupsterg is online now  
post #6 of 19 (permalink) Old 01-25-2017, 11:30 AM
Expert pin bender
 
dagget3450's Avatar
 
Join Date: Jul 2014
Posts: 1,878
Rep: 147 (Unique: 88)
I would be glad to help provide any data if it would be useful. I am going to have furyx on multiple platforms real soon. Going to get prepped for a lan party soon. biggrin.gif I can have furyx on x58/x79/x99/990fx - also would love to see FSU/FSE/FS and even Timespy runs. Curious how things like HBM overclocking affect 4k over 1080p

GPU i currently own: 390x/FuryX/Vega FE/RX Vega 64/1080TI - CPUs: 5960x/R7 1700/X5650x2/E5 2863/e5 2670
Radeon Vega Frontier Edition Owner
dagget3450 is offline  
post #7 of 19 (permalink) Old 01-25-2017, 12:20 PM - Thread Starter
Meddling user
 
gupsterg's Avatar
 
Join Date: Jan 2015
Location: Lurking over a keyboard
Posts: 6,931
Rep: 739 (Unique: 344)
FSE runs added to OP today smile.gif , FSU tomorrow mate smile.gif , TS soon wink.gif .

Feel free to do tests chap wink.gif , share whatever you wish to mate smile.gif .

The 2x Nitro cards I bought one was absolutely pants, +100mV to do 1100MHz, it was at 1350mV VID thumbsdownsmileyanim.gif . 2nd one needed 1275mV to gain 1100MHz stable. As I'd seen some members experiencing scaling issues even on custom PCB cards I knew I'd be wasting time playing with them frown.gif , so I RMA'd 1st and managed to make a small profit on selling 2nd smile.gif .

HBM OC makes as much of a scaling difference at 4K as 1080P, negligible IMO wink.gif . As I won't have created that section in OP for a few days, if you want a sneak peak at data, pick some GPU/HBM clocks shown in OP 3DM FS/FSE runs and I'll post like FSU ones smile.gif .
gupsterg is online now  
post #8 of 19 (permalink) Old 01-26-2017, 09:31 AM - Thread Starter
Meddling user
 
gupsterg's Avatar
 
Join Date: Jan 2015
Location: Lurking over a keyboard
Posts: 6,931
Rep: 739 (Unique: 344)
gupsterg is online now  
post #9 of 19 (permalink) Old 01-26-2017, 09:41 AM
Expert pin bender
 
dagget3450's Avatar
 
Join Date: Jul 2014
Posts: 1,878
Rep: 147 (Unique: 88)
Quote:
Originally Posted by gupsterg View Post

Today OP has been updated with 3DM FSU benchs, HML file / HWiNFO screenie plus some other info added in OP as well.

Nice!

On a side note i seem to be having wierd results on 390x firestrike. In firestrike if i raise vram from 1500 to 1700 i get lower combined scores. If i use 1700 on vram in FSE i get higher combined over 1500. Any ideas?

I havent got to fiji yet was trying to finish with 390x first since they are all custom blocked

GPU i currently own: 390x/FuryX/Vega FE/RX Vega 64/1080TI - CPUs: 5960x/R7 1700/X5650x2/E5 2863/e5 2670
Radeon Vega Frontier Edition Owner
dagget3450 is offline  
post #10 of 19 (permalink) Old 01-26-2017, 09:52 AM
Expert pin bender
 
dagget3450's Avatar
 
Join Date: Jul 2014
Posts: 1,878
Rep: 147 (Unique: 88)
Meant to add is it a speed/strap vs bandwidth possibly?

GPU i currently own: 390x/FuryX/Vega FE/RX Vega 64/1080TI - CPUs: 5960x/R7 1700/X5650x2/E5 2863/e5 2670
Radeon Vega Frontier Edition Owner
dagget3450 is offline  
Reply

Quick Reply
Message:
Options

Register Now

In order to be able to post messages on the Overclock.net - An Overclocking Community forums, you must first register.
Please enter your desired user name, your email address and other required details in the form below.
User Name:
If you do not want to register, fill this field only and the name will be used as user name for your post.
Password
Please enter a password for your user account. Note that passwords are case-sensitive.
Password:
Confirm Password:
Email Address
Please enter a valid email address for yourself.
Email Address:

Log-in



Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)
 
Thread Tools
Show Printable Version Show Printable Version
Email this Page Email this Page


Forum Jump: 

Posting Rules  
You may post new threads
You may post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off